Two-level multilevel model using SPSS (chapter 3 v1)

[@footballsailors2328]

Thank you for what you are doing. I appreciate. I wish if you author a book.

[@RuaTheRapoet]

Classic statistics video comment template: [gracious thank you for the content in the video, claiming it was all learned successfully], followed by [specific question that usually could be answered by the commenter with some diving into it solo, but that they want you to give them a quick solution for]

[@olgamolina2671]

Thanks for your kind tutorial! Could you please update the excel files links for generating plots? I wasn't able to download them :(

[@martamamyan5262]

Hello Mike, thanks a lot for the videos, they are really helpful. I have a question related to my data structure and the right methods to use with it. I am not interested in time trends, but I want to measure the effect of a country-level variable on an individual's probability to do something by taking advantage of using a bigger sample. My individuals are nested in 60 countries, and there are 2 time periods (2009 and 2015, but not all the countries are measured during both time periods and the individuals are not the same). When I run a multilevel model separately in time periods (one for 2009, another one for 2015), the results of my country-level variables seem logical and something that I would expect to receive but when I combine the two waves and add the year as a control variable in the multilevel model, the direction of my country-level variable changes (the coefficient becomes negative instead of being positive). Is this because I need to add the time variable as a level so that my time variable is my 1st level and the second block is the interaction of country and time? Thanks a lot in advance!

[@farnazmarianoshokrifard8446]

Can I do Multilevel nested model analysis on SPSS software when I don't have continuous variables?

[@suzanneyanney]

Thank you for these helpful and informative videos! I was wondering what to do if the null hypothesis test comes back as non-significant for the level 2 variable? Am I right in thinking that this suggests that the variance in dependent variable is related to individual/level 1 predictors and not level 2 predictors, and therefore a multilevel model may not be the appropriate test for the second level with predictors of the dependent variable?

[@mikecrowson2462]

Hi Suzanne, usually the way a non-significant variance component at L2 is taken as an indicator that there is no need to perform the multilevel model (and that you can go with a single level model). This might be a problematic way to go in situations where the slope(s) involving level 1 predictors (included in the model; not shown in this video) vary between Level 2 units even though the intercepts do not. But in my experience this is a fairly unusual circumstance.

[@suzanneyanney]

@@mikecrowson2462 That's really useful, thank you for your quick response!

[@tingtingzhang1442]

Thank you for all helpful videos! I really learned a lot. When I was running multilevel modeling with my data, I got an error message "The final Hessian matrix is not positive definite although all convergence criteria are satisfied. The MIXED procedure continues despite this warning. Validity of subsequent results cannot be ascertained." Can I ask whether any of your videos covers the reason for this error? Thank you again!

[@maxwelldapaah225]

Hi Mike, is it possible to do a hierarchical regression with fixed effects in SPSS? Will appreciate your thoughts or link to any videos you may have on this subject. I am working on testing the improvement or otherwise that occurs in health outcome x when you add two predictors (financial autonomy and accountability) to a first predictor (domestic government expenditure) in a panel data involving 39 countries over 15 years. I wanted to hold the country and/or year codes as fixed. Thanks

[@mikecrowson2462]

Hi Mawell, SPSS is kind of limited in terms of being able to analyze panel data using standard least-square regression. One option is to create dummy coded variables representing country membership (also the data should be in long format). Here's a video I put together awhile back on this approach (kzbin.info/www/bejne/a3Wtg2mAaK13gbc) Another option is to use generalized estimating equations with longitudinal data. SPSS does include that option. But I'll be honest, with 39 countries I probably just model the data using a multilevel model. Here's a video I put together recently (kzbin.info/www/bejne/ZmnIfayBna-Nqs0). It's a growth curve model, but there's no requirement you model growth curves (if you are not interested in trending over time). You could just include time-varying predictors of the outcome. Essentially, you would be allowing country to randomly vary (treated as randomly sampled from a population of countries) and then adding in time-varying predictors at Level 1. One thing though, is to consider centering the DV within countries so that you don't run into problems with bias as a result of between-country information filtering into the regression slope at Level 1. Here's a link to a paper I'm on where we have longitudinal data with school districts. It might give you an idea of how to proceed with your work (www.educationfinance.us/assets/Uploads/Maiden-Crowson-and-Byerly-Full-Report.pdf) . An article you might also find helpful is here (www.statmodel.com/download/HamakerMuthenFixedVSRandom.pdf). This is a white paper, but it is forthcoming in Psychological Methods. Best wishes!

[@ericnovotny7381]

Hello, thanks for the video. I did everything similar to you but got the error "The final Hessian matrix is not positive definite although all convergence criteria are satisfied. The MIXED procedure continues despite this warning. Validity of subsequent results cannot be ascertained." and "This covariance parameter is redundant. The test statistic and confidence interval cannot be computed." What can I do?

[@brittjane4486]

Hi Mike - just wondering if you have the slides for this series? I tried all the links but they don't work, probably because an old video. Thanks so much - I find MLM so hard to get my head around!

[@mikecrowson2462]

Hi Britt. Thanks for your question. You're right. That old site was deleted. I have the posted the link to the Powerpoint and data in one of my 'DEAR VIEWERS' comment. Also, see my other comment suggesting you visit a much newer (updated) video on HLM in SPSS. I hope you find these useful! Thanks for visiting!

[@brittjane4486]

@@mikecrowson2462 Hi Mike - thankyou for that. I can't find the comment with your powerpoint slides in it though?

[@mikecrowson2462]

@@brittjane4486 Hi. I have two separate comments below. The first one contains the Powerpoint and data links. But here's the link for the Powerpoint again: drive.google.com/open?id=1254QLr7mLKSjfznxe1FSqzY2AMg1QEDn

[@matildablicher-hansen3992]

Thanks for this! It is really clear. I am early in my MLM learning, and wondering whether this would work for n-of-1 - multiple observations within-person - with multiple predictor variables?

[@aparnasm9040]

Thank you so much sir.. for detailed explanation. It is really helpful. I am using HLM for my study and I was wondering whether we should perform model fit analysis (CFA)for multi-level models. If yes.. kindly give more insights or suggest some materials, sir.

[@julia7846]

Is there an approximate number & lower of level 2 units where you would recommend using REML instead of ML? Also thank you for your videos, they are extremely helpful! :)

[@jennifersweeney2792]

Hi Mike, This is really great and so useful as I'm learning MLM. In this video when you are conducting the regression model you refer to centering, is this group mean centering or grand mean centering? Many thanks in advance

[@mikecrowson2462]

Hi Jennifer. Yes, in the context of multilevel modeling you can use group-mean centering or grand-mean centering. FYI, I have a couple of newer videos that you might find interesting. Here's one that is a redo of the one you are commenting on: kzbin.info/www/bejne/rma9Zn6PhKmqhM0 Also, i have another here: kzbin.info/www/bejne/iIaUoaZsn8abm6s I believe I discuss centering in one or both (or at least there is talk of centering in some of the supplemental materials; sorry my memory fails me on where the discussion is) Hope this helps. Cheers!

[@jennifersweeney2792]

@@mikecrowson2462 thank-you, it does help. Essentially I'll use group mean at level 1 and grand mean at level 2. I've also ran the no predictors variance model and my covariance intercept has shown an error; 'The covariance parameter is redundant. The test statistic and confidence interval cannot be computed'. Does this tend to be a common issue that you've seen? I'm struggling to understand the cause through a Google search. Any help would be great, thank-you.

[@jennifersweeney2792]

I think it may have been because I didn't have enough members per team at the group level, so I amended my level 2 to look at the company sub-function and it worked. Guess that's because there was enough members in the team to produce variance. Only issue I have now is there are only 37 sub-functions.

[@oksaubercool]

You dont use the student ID for the random effects. I wonder if you could click in both the school ID and the student ID (as school*student) for the random intercept and would it make a difference?

[@mikecrowson2462]

Hi there. In HLM, the specification of random effects in the model pertain to the regression coefficients themselves. The intercepts in the intercept-only model are school means on the Level 1 outcome. The means for students' scores (computed for students within school) are presumed to vary BETWEEN schools. When we specify that the intercepts are allowed to vary, we are modeling variation between schools. If we were to have a predictor at Level 1, we might assume that each school would have a slope and that those slopes might vary between schools. Thus, we could specify the slope as a random coefficient in the model. So given that we are concerned with variation in slopes and intercepts across higher level units in HLM, it is no surprise it has also been referred to as random-coefficient modeling. So, no you do not want to include student ID as an identifier of the Level 1 units, since there are no within-student coefficients that can vary across students (of course, if you used repeated measures data, with repeated measurements nested within student - then student would become the Level 2 unit; but I digress here). I'll be honest. I assume if you included student ID as a Level 1 identifier in the SPSS specification, the program would either ignore it or give you an error message. Hope this helps!

[@oksaubercool]

Ah right, I was about repeated measurements here. Thanks a lot though. Could you ever do something about Generalized mixed models as well? I think your videos are quite helpful and stay 'grounded' in way which helps understanding.

[@mikecrowson2462]

Hi. I've already done several on multilevel binary logistic regression. You can go to this playlist to see them: kzbin.info/aero/PLITf5JBUH0l6PbUYl55oP4cbJC344cqKq

[@oksaubercool]

your playlist is so long, totally overlooked that, sorry ;-) and thanks so much

[@akpashtoon1981]

Hi dear Mike, can you please send me the power point slide that you are presenting in this video, it's quite informative. Thanking you in anticipation !

[@kristofferbo6977]

Thanks for this! It is really clear.

[@ninjasrini]

It will be really useful if you show the code. Thank you.

[@amiramohamed-hm9zq]

Thank you

[@mikecrowson2462]

DEAR VIEWERS: I have just recently uploaded a couple of new videos (as of July, 2019) that contain links to Powerpoints and the raw data underneath the video description: kzbin.info/www/bejne/iIaUoaZsn8abm6s and kzbin.info/www/bejne/hZysn56lobh-o5I . I also have reviewed HLM using a new program, called Jamovi: kzbin.info/www/bejne/mIimZ3-KZa9poKM . Please consider visiting these links, as they reflect my most current thinking on the topic!

[@sandeepbanger1]

unable to download ppts......nice explanation

[@mikecrowson2462]

DEAR VIEWERS: I HAVE JUST UPLOADED AN UPDATED (FEBRUARY 2020) REVIEW OF THE EXAMPLES PROVIDED IN CHAPTER 3 FROM HECK ET AL. (2014) CAN BE FOUND HERE: kzbin.info/www/bejne/rma9Zn6PhKmqhM0 . At that link you will also find a Powerpoint you can download, along with the data. Please check it out!

[@mikecrowson2462]

DEAR VIEWERS: THE POWERPOINT FOR THIS PRESENTATION CAN BE DOWNLOADED HERE: drive.google.com/open?id=1254QLr7mLKSjfznxe1FSqzY2AMg1QEDn . THE DATA CAN BE DOWNLOADED HERE: drive.google.com/open?id=1-u6z-LQ4ZyoWD6uMsAuFlMYt1LUKyVJa

[@afshinsh580]

Nice job

[@sunshine-yd3bd]

Thank you!